What jobs will AI replace? What the 2026 data actually shows

Between January and August 2026, US employers said artificial intelligence was the reason for 116,175 announced job cuts, according to the Challenger, Gray & Christmas August 2026 report. That makes AI the most cited reason for layoffs this year. It is also a smaller number than the headlines suggest, and it hides a more useful answer to the question you probably came here with.

Short answer: what jobs will AI replace?

AI is most likely to replace jobs, or large parts of jobs, that are built mainly from reading, writing, sorting and answering at a screen: customer service, data entry, translation, routine programming, proofreading, and clerical and admin work. The jobs least likely to be replaced by AI are physical, hands-on or deeply personal ones, such as roofers, electricians, nurses' aides, cooks and massage therapists. So far, AI has mostly changed tasks rather than eliminated whole occupations, and the clearest damage is to hiring for junior roles.

That is the summary. The rest of this piece explains where those lists come from, what the evidence actually shows as of late October 2026, and how you can judge your own job. If you want the flip side (the new jobs AI is creating and what they pay), see our guide to AI jobs in 2026.

How it works: AI replaces tasks before it replaces jobs

Here is a question worth asking before you panic about any ranking: what exactly is being measured? Almost every "jobs AI will replace" list is built the same way. Researchers break each occupation into tasks, using the US government's O*NET job database, and then ask how many of those tasks an AI system can do or help with. The result is an exposure score, not a forecast of layoffs.

Think of it like a kitchen getting a dishwasher. The appliance takes over one task completely. It does not replace the cook, but it might mean a restaurant needs one fewer kitchen porter. Whether that porter loses their job depends on things the dishwasher knows nothing about: how busy the restaurant is, how cheap labour is, and whether the owner would rather grow than cut.

The two most-cited recent studies show how this works. In July 2025, Microsoft researchers analysed 200,000 anonymised Bing Copilot conversations and scored occupations by how much their work activities overlapped with what people actually used AI for. Interpreters and translators came top, followed by historians, passenger attendants, services sales representatives, writers and customer service representatives. The researchers were careful to say they had "measured how AI capabilities overlap with work activities," not job losses.

In March 2026, Anthropic published a measure it calls "observed exposure", which combines what large language models can theoretically do with real usage data from Claude and gives more weight to uses that automate work than to uses that assist it. Its top ten most exposed occupations were computer programmers, customer service representatives, data entry keyers, medical record specialists, market research analysts, wholesale sales reps, financial analysts, software testers, information security analysts and computer support specialists. About 30% of workers had zero coverage at all.

Put the two studies side by side and a clear pattern emerges.

Most exposed to AI	Least exposed to AI
Customer service representatives	Roofers and cement masons
Interpreters and translators	Water treatment plant operators
Writers, editors and proofreaders	Surgical assistants and orderlies
Computer programmers and software testers	Massage therapists
Data entry keyers and medical record specialists	Cooks, bartenders and dishwashers
Market research and financial analysts	Motorcycle mechanics and logging equipment operators

Sources: Microsoft Research via GeekWire and Anthropic via Repairer Driven News. The table groups occupations named in each study's lists; it is not a single combined ranking.

One finding surprises people. The Anthropic data shows the most exposed workers are more likely to be educated and well paid: they earn 47% more on average, and graduate degree holders make up 17.4% of that group compared with 4.5% of the unexposed group. Earlier waves of automation hit factory floors first. This one starts at the desk.

Real examples: what jobs are being replaced by AI so far?

Exposure is one thing. What has actually happened? The honest answer is less than the lists imply, with a few sharp exceptions.

Customer service is the clearest test case, and it cuts both ways. In early 2024, Klarna said its AI assistant was handling 75% of customer chats and doing the work of 700 agents. By May 2025, its CEO told Bloomberg the chatbot's output was "lower quality" and the company started recruiting human agents again, promising "there will always be a human if you want." AI could do the volume. It struggled with the cases where customers most needed help.

Young workers are where the effect shows up most clearly in the data. A Stanford Digital Economy Lab paper by Erik Brynjolfsson and colleagues, using ADP payroll records, found that workers aged 22 to 25 in the most AI-exposed occupations saw a 13% relative decline in employment, while older workers in the same jobs were stable or growing. Customer service, accounting, software development and admin roles were hit hardest. We cover that entry-level story in depth in AI agents at work: what it means for jobs.

Across the whole economy, though, the picture is calmer. The Yale Budget Lab found no major shift in the mix of occupations since ChatGPT launched and no change in how long workers in exposed jobs stayed unemployed. Federal Reserve Bank of Chicago research reported that employment in the most AI-exposed occupations actually rose 4.05% from 2019 to 2024, and Anthropic's own study found "no systematic increase in unemployment" for highly exposed workers since late 2022.

What about those layoff numbers? AI was blamed for about 22% of announced US cuts so far in 2026, per Challenger. In 2025 the share was 4.5% over the first 11 months, according to Oxford Economics, which suspected some firms were "trying to dress up layoffs as a good news story." Yale's Martha Gimbel made a similar point to Fortune: blaming AI can be easier than admitting to over-hiring or tariff pain. So part of the rise is real, and part is probably relabelling. Nobody can yet say how much of each.

Benefits and limitations of the "AI will take your job" lists

These rankings are useful. They show which kinds of work are shifting first, and they are based on real usage rather than guesswork. If your job sits near the top of both the Microsoft and Anthropic lists, that is a signal worth taking seriously.

But they have three blind spots. First, exposure counts tasks, not jobs. A financial analyst whose spreadsheet work gets faster may simply analyse more companies. The Microsoft researchers themselves pointed to bank tellers and ATMs: the machines took over cash handling, and the job changed rather than vanished. Second, the data comes from chatbot logs, which over-represent people who already use AI and say little about whether a company has rebuilt its workflow around it. Third, the lists ignore cost, regulation and trust. A model might draft a medical record summary, but someone still has to sign it.

The forward-looking forecasts carry the same uncertainty. The World Economic Forum's Future of Jobs Report 2025 expects 170 million jobs to be created and 92 million displaced by 2030, a net gain of 78 million. Its fastest-declining roles are cashiers and ticket clerks, administrative assistants, printing workers, and accountants and auditors. Its fastest-growing roles in absolute numbers include farmworkers, delivery drivers, construction workers, nurses and care workers. That is a survey of employers' expectations, not a measurement, but it lines up with the exposure data.

What it means for you: a simple test for AI-proof jobs

No job is completely AI-proof, but some are far more resistant than others. Rather than hunting for a magic list of AI-proof careers, it helps to score your own work. We put together this rough test from the patterns in the studies above. It is a thinking tool, not a model, so treat the answers as a direction rather than a verdict.

Question about your job	More exposed if...	More resistant if...
Where does the work happen?	Almost entirely on a screen	In the physical world, with tools or bodies
What is the output?	Text, code, data or answers	A repaired object, a treated patient, a built thing
How standard is each case?	Similar requests, clear rules	Every case is different and messy
Who carries the risk?	Mistakes are cheap and easy to fix	A licensed person must sign off and be accountable
How much is relationship?	Customers rarely care who did it	Trust, persuasion or care is the product

If most of your answers fall in the middle column, the practical move is not to flee your field but to shift your weight within it. Do more of the judgement, client and sign-off work, and get good at directing AI on the routine parts. People who can check an AI's output are worth more than people who compete with it on speed. Short, practical AI training courses are one low-cost way to start, and it is worth learning how agentic AI differs from a chatbot, because agents are what will take over multi-step tasks next.

If you are early in your career, the data says to be most careful. Entry-level hiring in exposed fields is where the pressure shows first, so internships, apprenticeships and visible portfolio work matter more than they did five years ago.

What happens next

Here we should separate what we know from what we are guessing. We know AI is already doing a meaningful share of tasks in customer support, coding and writing. We know younger workers in those fields are being hired less. We do not know whether this becomes broad job loss or a slow reshuffle like the one that followed the personal computer.

Gimbel's view, reported by Fortune, is that a recession would be the real test, because tight budgets push firms to automate faster. That seems right to us. Past technologies took decades to reshape work, not months, and the current data still shows more change inside jobs than disappearance of them. Our best read is that the jobs AI will replace first are the ones made of repeatable screen tasks, and that the people in them have time to move up the value chain, but not unlimited time. Rules on how AI can be used in hiring and firing are also still being written; our AI regulation guide tracks where that stands.

FAQ

What jobs are safe from AI?

The jobs safest from AI combine physical work, unpredictable settings and human trust. Microsoft's and Anthropic's studies both rank skilled trades (roofers, cement masons, mechanics), hands-on healthcare (surgical assistants, orderlies, massage therapists), and food service roles like cooks and dishwashers among the least exposed. Jobs that need a licensed person to take responsibility are also more protected.

Is AI taking over jobs?

AI is taking over tasks faster than whole jobs. Employers blamed AI for about 22% of announced US job cuts in 2026 through August, per Challenger, but the Yale Budget Lab and the Chicago Fed find no broad disruption yet. The clearest impact is fewer entry-level hires in exposed fields, with a 13% relative employment drop for 22 to 25-year-olds.

What jobs will AI not replace?

AI is unlikely to replace jobs where the core value is physical skill, care or accountability: plumbers, electricians, nurses, care workers, construction trades, firefighters and therapists. It will still change these jobs at the edges, for example through paperwork, scheduling and diagnostics, but the central work needs a person who is physically there and trusted.

When will AI take over?

Nobody knows, and anyone giving a precise date is guessing. The WEF expects large shifts by 2030, with more jobs created than displaced. Economists at Yale note that earlier technologies such as computers reshaped work over decades. A recession could speed adoption. The most likely path is gradual change, job by job, rather than a single takeover moment.

What jobs are being replaced by AI?

The jobs being replaced by AI most visibly today are customer service agents, data entry clerks, translators, copywriters, proofreaders and some junior programming and admin roles. Even here, replacement is partial: companies like Klarna cut agents for chatbots and then rehired humans when quality fell. Most affected workers see their tasks shrink rather than their job disappear overnight.